3

Please would you help me with this question? I've been thinking about it for ages but I've made very little headway, so if possible a hint would be ideal.

Let $\sum_{n=1}^∞{x_n}$ be a divergent series, where $x_n > 0$ for all $n$. Show that there is a divergent series $\sum_{n=1}^∞{y_n}$ with $y_n > 0$ for all $n$, such that $(\frac{y_n}{x_n}) → 0.$

I have not been taught analysis formally, hence my lack of progress. I know to consider the series as a sequence of partial sums, and I tried to take the contrapositive of the statement but that just overcomplicated matters. I know I don't have many ideas to present but I have been trying this for days.

Thank you in advance.

  • 1
    HINT: start by looking at some concrete examples. For instance, consider $x_n:=1/n$ and start exploring from here – Joe Sep 26 '17 at 21:13

4 Answers4

1

You can break up your divergent series $\sum x_n$ into bunches of consecutive terms, each of which are $>1$. In detail there are $n_1$, $n_2,\ldots$ with $$x_1+x_2+\cdots +x_{n_1}>1,$$ $$x_{n_1+1}+x_{n_1+2}+\cdots +x_{n_2}>1,$$ etc. Pick suitable constants $u_1,u_2,\ldots$ and set $y_k=u_ix_k$ whenever $n_{i-1}<k\le n_i$.

Angina Seng
  • 158,341
1

There are some 'clever' ways to do this, but my favourite way falls into the 'just do it' category.

We look for a sequence of the form

$$ y_n=a_nx_n $$

where $a_n\to 0$. To simplify things, let's let $(a_n)$ be of the form:

$$ \underbrace{1,\cdots,1}_{N_1},\underbrace{\frac{1}{2},\cdots,\frac12}_{N_2},\underbrace{\frac13,\cdots,\frac13}_{N_3},\underbrace{\frac14,\cdots\frac14}_{N_4}\cdots $$

for suitable constants $N_1,N_2,\cdots$.

Now, since we know that $\sum_{n=1}^\infty x_n$ is divergent, we can pick $N_1,N_2,\cdots$ to be such that

\begin{gather} x_1+\cdots + x_{N_1} \ge 1 \\ x_{N_1 + 1} + \cdots + x_{N_2} \ge 2 \\ x_{N_2 + 1} + \cdots + x_{N_3} \ge 3 \end{gather} and so on. With such a choice of $N_1,N_2,\cdots$, we see that\begin{gather} y_1+\cdots + y_{N_1} \ge 1 \\ y_{N_1 + 1} + \cdots + y_{N_2} \ge 1 \\ y_{N_2 + 1} + \cdots + y_{N_3} \ge 1 \end{gather} and so $\sum_{n=1}^\infty y_n$ is divergent.

John Gowers
  • 24,959
1

Based on this result:

If the positive series $\sum a_n$ diverges and $s_n=\sum\limits_{k\leqslant n}a_k$ then $\sum \frac{a_n}{s_n}$ diverges as well

Let's take $S_n=\sum\limits_{k=1}^n x_n\to+\infty\quad$ then $\quad\displaystyle y_n=\frac{x_n}{S_n}$ agrees with your requirements.

  • $x_n>0\implies S_n>0\implies y_n>0$

  • $\displaystyle \frac{y_n}{x_n}=\frac 1{S_n}\to 0$

  • $\sum\ y_n$ diverges


For instance the classical divergent series $x_n=\frac 1n\ $ gives $\  y_n\sim\frac 1{n\ln(n)}$

zwim
  • 28,563
-1

To make it even easier:

Define $(y_{n})$ as $y_{1}=\frac{x_{1}}{1}, y_{2}=y_{3}=\frac{x_{2}}{2}, y_{4}=y_{5}=y_{6}=\frac{x_{3}}{3},y_{7}=y_{8}=y_{9}=y_{10}=\frac{x_{4}}{4}\dots$ where every time $\frac{x_{n}}{n}$ $n$ is repeated $n$ times...

Julien
  • 363
  • No. $(y_{n})$ will be $1, \frac{1}{8}, \frac{1}{8}, \frac{1}{12}, \frac{1}{12}, \frac{1}{12}, \frac{1}{36}, \frac{1}{36}, \frac{1}{36}, \frac{1}{36}, \dots...$ – Julien Sep 26 '17 at 21:59
  • $y_{4} = \frac{x_{3}}{3} = \frac{1}{12} \neq \frac{1}{27}$. – Julien Sep 27 '17 at 07:39
  • I apologise; I made a mistake. Did you come up with this yourself or did you see it somewhere? It's much simpler than all the 'standard' answers I've heard, which is why I was convinced it had to be wrong. If you edit the answer then I'll change the downvote to an upvote. – John Gowers Sep 27 '17 at 14:05
  • 2
    No problem, the indices were confusing as well. I found it myself by realizing that in order to guarantee divergence an easy way might be to "spread" the partial sums over a larger span of indices – Julien Sep 27 '17 at 18:31
  • I'm not convinced yet, though, that $\frac{y_n}{x_n}\to 0$. – John Gowers Sep 27 '17 at 20:23
  • Indeed, let $x_\frac{n(n+1)}{2}=\frac{1}{n}$ for all $n$ and let $x_m=1$ otherwise. By your definition, we have $y_{\frac{n(n+1)}{2}}=\frac{x_n}{n}$. Now let $m$ be some number not of the form $\frac{n(n+1)}{2}$. Then $y_{\frac{m(m+1)}{2}} / x_{\frac{m(m+1)}{2}}$ is given by $\frac{1}{m} / \frac{1}{m} = 1$. But there are infinitely many such numbers $m$ and so the sequence $(y_n/x_n)$ has a subsequence that does not converge to $0$ and therefore does not converge to $0$ itself. – John Gowers Sep 27 '17 at 20:36
  • Not checking the details but I get the idea - you are correct! My error. – Julien Sep 27 '17 at 20:51